Abstract:
The objective of classifcation is to assign a class to a given data instance. One
well-recognized classifer is the k-NN model, where the class of an instance is determined
by considering the majority class among its k nearest neighbors. However, k-NN’s performance weakens in imbalanced datasets. To address this, adjusting k for each instance based on factors like its position
relative to clusters or isolation, and integrating density-based scores from a parameterfree Mass-ratio-variance Outlier Factor (MOF) into the k-NN process, helps determine
suitable nearest neighbors.
Our research focuses on the development of a dynamic nearest neighbor classifer
tailored specifcally to address class imbalance problems. Experimental results on ten
real-world datasets show our classifer accurately forecasts outcomes, aligning closely with
traditional k-NN with the best parameter k.
